Semantic similarity assessment of words using weighted WordNet

摘要

Word and concept similarity assessment is one of the most important elements in natural language processing and information and knowledge retrieval. Word-Net, as a popular concept hierarchy, is used in many such applications. Similarity of words in WordNet is also considered in recent researches. Many researches that use WordNet, have calculated similarity between each pair-word by considering depth of subsumer of the words and shortest path between them. In this paper, three novel models to make better semantic word similarity measure have been presented and it was improved by giving weights to the edges of WordNet hierarchy. It was considered that the nearer an edge is to the root in the hierarchy, the less effect it has in calculating the similarity. Therefore, we have offered a new formula for weighting the edges of hierarchy and based on that, we calculated the distance between two words and depth of words; and then tuned parameters of the transfer functions using particle swarm optimization. Experimental results on a common benchmark, created by human judgment, show that the resultant correlation improved; furthermore our formulae were applied to a more realistic application called sentence similarity assessment and it led to the better results.